Low accuracy levels are often obtained when readers are asked to predict test performance over reading materials. Three investigations further explore the information readers use to make predictions during metacomprehension. Our results show that readers' estimates are influenced by factors such as their initial impression of the reading task, based in part on their perceptions surrounding text genre and test item type. To explain these and other published results, a new framework for investigating metacomprehension using Tversky and Kahneman's (Science, 185:1124-1131, 1974) anchoring and adjustment heuristic as a guide is proposed. We argue that readers anchor comprehension test performance on factors such as self-perceptions of reading ability and/or perceptions of the reading task and then insufficiently adjust their predictions to reflect the demands of the specific reading task at hand such as text difficulty.